A Preliminary Study on Symbolic Fuzzy Cognitive Maps for Pattern Classification

Within the neural computing field, Fuzzy Cognitive Maps (FCMs) are attractive simulation tools to model dynamic systems by means of well-defined neural concepts and causal relationships, thus equipping the network with interpretability features. However, such components are normally described by quantitative terms, which may be difficult to handle by experts. Recently, we proposed a symbolic FCM scheme (termed FCM-TFN) in which both weights and activation values are described by triangular fuzzy numbers. In spite of the promising results, the model's performance in solving prediction problems remains uncertain. In this paper, we explore the prediction capabilities of the FCM-TFN model in pattern classification and concluded that our method is able to perform well when compared with traditional classifiers.